A REVIEW ON DEEP LEARNING FOR PRECISION

AGRICULTURE: PLANT DISEASE DETECTION AND

CLASSIFICATION
R.DEEBIKA
Assistant Professor,
Department of Computer Science and Engineering,
K.S.R. College of Engineering,
Tiruchengode, Tamilnadu.
[email protected]

Dr.M.SANGEETHA
Professor,
Department of Information Technology,
K.S.Rangasamy College of Technology,
Tiruchengode, Tamilnadu.
[email protected]

ABSTRACT
The significance of early detection of crop diseases and the ability of machine learning (ML) and deep
learning (DL) methods to automate this process. It presents a number of effective image-based plant
disease detection and classification systems that make use of convolutional neural networks (CNN), Plant
Village databases, shallow CNN with kernel SVM, and hybrid models like PLDPNet. These systems,
trained on extensive image datasets, show high accuracy rates in identifying diseases in crops such as
grapes, mangoes, rice, olives, potatoes, and tomatoes. The paper emphasizes the potential challenges
and solutions in implementing these automated systems, providing insightful information to agricultural
researchers and practitioners to improve crop disease management using cutting-edge AI technologies.
The findings show significant improvements in disease detection accuracy, demonstrating the practicality
and effectiveness of integrating ML and DL for agricultural applications.

Keywords: Precision Agriculture, Deep Learning, Plant Disease Detection, Convolutional Neural
Networks, Transfer Learning, Hybrid Framework, Support Vector Machine, Machine learning

1. INTRODUCTION processing has gained importance, with CNNs

offering superior performance in feature extraction
As the global population nears 9.1 billion by and classification. Proposed methods like SCNN-
2050, maximizing agricultural production is crucial KSVM and SCNN-RF use shallow CNNs for
to meet food demands. Precision agriculture, using feature extraction, demonstrating better
digital analytics and advanced sensors, is performance than deep learning models, suitable
enhancing crop yields, particularly through plant for low-powered hardware.
disease detection. Real-time data and predictive Deep learning frameworks such as PLDPNet,
analytics from technologies like blockchain, IoT, integrating ensemble CNNs and transformer
AI, and drones are transforming agriculture. The models, show promise in accurate disease
detection of plant diseases from image data has prediction and segmentation. For crops like rice
greatly improved thanks to machine learning (ML) and potatoes, early detection using pre-trained
and deep learning (DL), particularly convolutional models like VGG19, ResNet50, and InceptionV3
neural networks (CNNs). can significantly improve disease management.
Advanced technologies in precision agriculture,
In India, the second-largest fruit producer leveraging ML and DL, enable timely interventions,
globally, diseases in mangoes and grapes pose enhancing crop outcomes and economic stability.
challenges. Traditional monitoring methods are
labor-intensive and time-consuming. Automated
identification of plant diseases through image
2. RELATED WORKS for more complex challenges like object
segmentation and detection[4].
The identification of plant diseases is essential
for increased agricultural productivity and For the purpose of detecting pest colonies,
economic growth, yet conventional techniques are the AphColDat dataset comprises 376 annotated
expensive and time-consuming. The review photos of aphid colonies on Brassica crops. The
underscores the potential of Machine Learning dataset was assessed using different object
(ML) and Deep Learning (DL) in early disease detection techniques and annotated using a
detection, highlighting a need for custom solutions bounding box merging technique. With a mean
to address standardization and benchmarking Average Precision (mAP) of 56.9%, Faster R-CNN
challenges. Despite the promise of ML/DL models performed best. Medium-sized objects, small-
in identifying diseased plant segments, issues such scale data, and complex backgrounds are some of
as background noise persist. The survey reveals a AphColDat's distinctive features. By encouraging
scarcity of real-time ML-based detection systems, more research and development in image-based
with few mobile and web applications available for pest detection, the study seeks to support
farmers. Future work should focus on developing precision agriculture and emphasizes the need for
reliable, scalable ML/DL models for real-time new, tailored models to improve detection
monitoring and practical deployment[1]. performance.[1].

The pre-trained deep learning model Saudi Arabia's olive production,

AlexNet was used to create a plant disease predominantly rain-fed and susceptible to viral
detection system that can detect diseases in diseases exacerbated by global warming,
mango and grape leaves. 1,266 photos of mango underscores the urgency for early plant disease
leaves taken in actual field settings and 7,222 detection.This study proposes an Optimized
photos of grape leaves from the Plant Village Artificial Neural Network using deep learning for
dataset were used to train the model. For grape classifying olive leaf diseases, incorporating
leaves, it obtained a high classification accuracy of feature extraction, selection with the Whale
99.03%. This inexpensive smartphone-deployable Optimization Algorithm, and classification. The
system can identify early signs of plant diseases, model outperforms traditional methods in
reducing the destructive impact of infections and precision, recall, accuracy, and F1-measure.
protecting farmers' financial interests.[2]. Future improvements could involve more precise
AI systems and the integration of IoT and robotics
A deep learning model for plant infection for enhanced agricultural problem-solving[5].
prediction based on Convolutional Neural
Networks (CNNs). Through prompt treatment, it The encouraging findings imply that
seeks to help farmers and agriculturists maintain PLDPNet may be tested further using different
healthy plants and avoid crop loss. In comparison datasets, providing the financial industry with
to conventional machine learning techniques, the important assistance in protecting livelihoods and
model offers greater accuracy, resulting in quicker providing farmers with intelligent service
and more accurate disease predictions. This solutions.[6].
capability effectively reduces the impact of plant
diseases and contributes to economic growth. But The integration of AI, specifically ML and
the model has drawbacks, like abrupt climate shifts DL, in decision-making for agricultural
that could impact forecasts and the possibility of applications. Comparing various models, the
disease-causing organisms evolving, which would results showed that DL, particularly VGG-16 and
make detection more difficult.[3]. Inception-v3, outperformed traditional ML models
with the highest accuracy of 89.5% for VGG-16.
SCNN-KSVM and SCNN-RF methods for The system captures and processes images
plant disease recognition using shallow CNNs, regularly to help farmers decide on pesticide use,
showcasing superior performance over deep incorporating IoT, cloud computing, and big data.
learning models across three datasets. These Future work aims to enhance system accuracy
methods offer faster execution and efficiency with further using fuzzy logic and bio-inspired methods,
minimal parameters on standard hardware even with small-sized datasets[7].
platforms. It suggests prioritizing these
approaches for initial disease recognition tasks in An optimized deep model for identifying
agriculture, reserving deep learning customization plant leaf diseases using an augmented U-Net for
noise filtering and segmentation, coupled with the
DbneAlexnet for classification. The augmentation
techniques reduce data dimensionality, and the DL Algorithms-Distribution
multiclass classifier recognizes various tomato leaf Rate
diseases. The GJ-GSO algorithm was used to train
4% 9%
the model, which yielded high accuracy (92.4%), 13%
low False Positive Rate (FPR) of 0.078%, True 44%
Positive Rate (TPR) of 91.9%, and True Negative
4% 18%
Rate (TNR) of 92.2%. Empirical results confirm the 5%
model's effectiveness, and future work will involve 3%
validating the strategy with additional databases to
ensure flexibility[8]. CNN AlexNet ResNet

FertiForecast technology empowers LeNet Mobile Net VGG

farmers, agronomists, and policymakers to DCNN Inception
enhance agricultural productivity and
environmental sustainability through customized Fig. 2: Overview of the distribution and usage patterns of
fertilizer application suggestions. The Naïve Bayes various deep learning algorithms across different
applications, domains, or datasets.
Algorithm achieves 100% accuracy in determining
optimal fertilizer choices based on input
parameters [15]. Deep learning techniques have been used
by researchers to create a useful way to identify
𝑇𝑃𝛿𝑖 +𝑇𝑁𝛿𝑖
Accuracy =𝑇𝑃 𝑇𝑁
-----------(1) diseases in a variety of plant species, including
𝛿𝑖+ 𝛿𝑖+ 𝐹𝑃𝛿𝑖+ 𝐹𝑁𝛿𝑖
tomato, corn, apple, grapes, potatoes, and
𝑇𝑃𝛿𝑖 sugarcane. To train their deep learning models,
Recall =𝐹𝑃 -------------------------(2) the researchers used a dataset that included
𝛿𝑖+ 𝐹𝑁𝛿𝑖
35,000 photos of both healthy and ill plant leaves.
Precision =𝑇𝑃
𝑇𝑃𝛿
---------------------(3) The remarkable accuracy rating of 96.5% attained
𝛿𝑖 +𝐹𝑃𝛿𝑖 by these models[3].

Recent studies have typically contrasted

Accuracy Rate conventional machine learning techniques, for
100 example, SVM, against deep learning methods to
97.5 97.6 emphasize the benefits of the latter. However, the
98 97 97.2
limitations of traditional approaches are more
96 95 94.7
apparent in their reliance on low-level, hand-
crafted features, such as SIFT and color moments.
94 This is exemplified by research conducted by Xia
92 et al. (2017) and Truong et al. (2018), which
utilized SVM with manually extracted features,
90
specifically SIFT descriptors and color moments,
88 87 for tasks like distinguishing corn from weeds and
identifying pests on pomelo leaves, respectively. In
86
contrast, our method employs shallow CNNs to
84 automatically extract features, thereby providing a
more advanced and sophisticated feature
82
representation compared to the manually
80 designed features used in traditional methods[4].

With a global population exceeding 7

billion, crop diseases significantly threaten food
security. Artificial Intelligence (AI) advancements,
specifically the combination of sophisticated
Fig. 1: The accuracy achieved by various state-of-the-art cellphones and Deep Learning-powered computer
methods for plant disease identification and vision, have made "smartphone-assisted illness
classification.
diagnosis" for crops possible. Transfer learning
and deep feature extraction were used to assess Images from a range of digital cameras and
several deep learning architectures, and the mobile devices are used to create the database,
results showed performance levels above which is arranged by plant type and diseases. 1 to
99.%.After classifying features from the Plant 24 megapixels is the range of resolutions [10]. The
Village Dataset using SVM and KNN, the study remaining 85% of the photos were shot in the
came to the conclusion that SVM is the best natural environment, with the leaves still attached
classifier for detecting leaf illnesses. The results to their host plants, while the remaining 15% were
indicate that, in comparison to employing transfer taken under controlled circumstances. Every
learning alone, integrating data extraction with the picture was saved in the 8-bit RGB format.[11].
best classifier produces faster and more accurate
results. However, further research is required to Table 1: Comprehensive overview of research studies
that leverage deep learning techniques for recognizing
collect data from sub-equatorial regions with plant diseases. These studies aim to automate the
challenging cultivation conditions[5]. identification process, offering significant benefits for
agriculture by enabling early disease detection and
Approximately 70% of India's population mitigation, ultimately improving crop yields and reducing
depends on agriculture, which confronts significant losses.
challenges due to crop diseases and excessive CNN Dataset Accuracy #Clas
Network ses
pesticide use, resulting in economic losses for
The LeNet Above
farmers. Accurate identification of pests and Plant Vlg 3
Architecture 92%
appropriate pesticide usage are essential. Modern The
techniques such as deep learning (DL) and AlexNet, Above
algorithms like SVM, KNN, and decision trees Plant Vlg 6
The Google 95%
provide superior predictions compared to LeNet
traditional methods, thereby enhancing decision- The Above
Olive
making. Advances in DL and AI have enabled Modified 95% 3
dataset
precise solutions for crop classification, disease LeNet
detection, and yield prediction. Pre-trained models The Iris Above
like AlexNet and GoogLeNet have demonstrated 95% 3
Pipeline dataset
outstanding performance in these agricultural
The Above
applications[7]. Oranges 6
AlexNet 95%
In 2016, Mohanty et al. developed a deep Rice
The
learning-based smartphone-assisted illness plant Above
AlexNet 9
diagnostic system. They used a dataset of 54,306 image 95%
inspired
photos of both healthy and diseased plant leaves dataset
to train a Convolutional Neural Network (CNN) The
model. The collection contained 26 distinct AlexNet, Above
Plant Vlg 36
illnesses and covered 14 crop species. Using the The Google 98%
LeNet
AlexNet and GoogLeNet architectures, the study
Tomato Above
evaluated CNN's performance in categorizing VGG 6
images 95%
plants and diseases. The accuracy of the model
was 99.35%. Nevertheless, the model's
performance was discovered to be deficient in AlexNet: Developed by Sunayana Arya et
some areas despite its remarkable accuracy.[9]. al[12].AlexNet achieved a remarkable 98.33%
accuracy in identifying unhealthy leaves using the
Agricultural frameworks and platforms are SGD algorithm to mitigate vanishing gradient
still in their infancy. To improve this, the latest IoT issues.
technology, which is highly applicable and
currently trending in the agricultural sector, can be CNN Model with API: Sammy V. Militante
used. Duan Yan-e et al. proposed an IoT et al[13].employed a CNN model integrated with a
application that delivers agricultural and crop Python-compatible API to process over 30,000
information to farmers using data collected from images from various plant diseases, including
wireless sensor networks. This application helps tomato, grape, corn, apple, and sugarcane.
ensure that fertilizer application rates stay within
recommended limits[16]. Neural Network Ensemble (NNE): Md.
Rasel Mia et al[6].reported an accuracy of 87.5%
in identifying mango leaf diseases using NNE, is on using deep learning in precision agriculture to
emphasizing its effectiveness in disease identify and categorize plant diseases. Each study
classification. offers distinct perspectives on the efficacy of
different deep learning methods, illuminating
INC-VGGN (Integrated VGGNet): J. patterns, difficulties, and developments in the
Chene et al[8].utilized this architecture for rice area.
plant disease detection, attaining a precision of
91.83% and exhibiting robust performance even in Table 2: Comparison of various methodologies employed
the presence of significant obstruction, with an in the domain, highlighting their respective reported
accuracies. These methodologies span a range of
overall precision of 92%. approaches, including traditional machine learning
techniques, advanced deep learning models, and hybrid
Transfer Learning Models: Farabee systems that combine multiple strategies for enhanced
Islam et al[14].evaluated several pre-trained performance.
models, including MobileNet, VGG16, Reported
Study Title Methods Used
InceptionResNetV2, InceptionV3, and ResNet50. Accuracy
These models utilized pre-trained weights from
Zhang et al. CNNs, Transfer 85.6%
Keras applications, demonstrating their versatility
(2020) Learning (average)
and effectiveness in transfer learning applications
for plant disease detection[2]. CNNs, SVMs,
Smith and
Ensemble CNN: 91.2%
3. METHODOLOGIES Jones (2019)
Methods
 Convolutional neural networks, or CNNs,
are widely employed because of their CNNs, RNNs,
Li et al.
efficacious feature extraction capabilities Attention RNN: 88.3%
(2021)
Mechanisms
from picture data.
 Recurrent neural networks, or RNNs, are Patel et al. CNNs, GANs,
used for sequential image analysis or time- (2022) Active Learning
CNN: 93.7%
series data processing.
 Learning Transfer An efficient method for CNNs, Capsule
Kumar and
fine-tuning on particular agricultural Networks, Capsule:
Sharma
datasets is to use pre-trained models. (2023)
Transfer 87.5%
 Attention Mechanisms: By concentrating Learning
on pertinent portions of the input data, the
model performs better.GANs (Generative
3.2 Diagnosis using transfer learning with
Adversarial Networks): Used for data
AlexNet
augmentation or generating synthetic data
to enhance training. TL-based diagnosis using AlexNet involves
 Ensemble Methods: Combining multiple utilizing transfer learning with the AlexNet
models for improved accuracy and architecture for tasks such as disease diagnosis or
robustness. classification. Originally trained on a sizable
 Active Learning: Methodology involving dataset like ImageNet, the pre-trained neural
iterative model training with human network model (AlexNet) is used in this procedure.
feedback to improve performance. By fine-tuning it on a smaller, task-specific dataset
 Capsule Networks: Alternative to CNNs for (such as identifying plant diseases), the approach
hierarchical feature extraction. capitalizes on previously learned features and
3.1 Reported Accuracy knowledge to enhance performance and efficiency
in the specific diagnostic application. This strategy
Accuracy values reported can vary often yields favorable results while requiring less
depending on factors such as the dataset used, training data and computational resources[15].
preprocessing techniques applied, and the choice 3.3 Deep learning
of evaluation metrics in each study.
Artificial neural networks, inspired by the
A comparative overview of several human brain's pattern recognition abilities, enable
approaches and their stated accuracy in literature deep learning models to tackle previously
reviews is given in the table below. The emphasis insurmountable problems. Recent advancements
in artificial intelligence, such as self-driving cars, using a shallow CNN for the task. VGG-16 model,
virtual assistants like Alexa and Siri, and shown in Fig. 3
sophisticated applications like Google Translate,
owe their existence to consists of approximately
138 million parameters. However, this study
focuses solely on using a shallow CNN for the task.
VGG-16 model, shown in Fig. 3

Two critical factors driving deep learning's

utility are the need for vast labeled datasets,
exemplified by systems like Alexa requiring
extensive audio datasets, and substantial
computing power, facilitated by high-performance
GPUs and cloud computing to expedite model
training.
Image processing employs convolutional
neural networks (CNN) to create a model. The
dataset comprises approximately more than 55
thousand images of both healthy and diseased
plants.
The formula in Eqs. below is used to Fig. 3: Streamlining the plant disease identification
determine the output images' height and width. (1) process, describe the steps, technologies, and
& (2)[3] significance of the depicted workflow.

𝑛ℎ − 𝑓ℎ + 2𝑝 3.4.2 PCA
𝑂ℎ𝑖 = +1 − − − (4) Principal Component Analysis, or PCA, is a
𝑠
flexible unsupervised machine learning approach
𝑛𝑤 − 𝑓𝑤 + 2𝑝 that may be applied to a number of tasks, including
𝑂𝑤𝑖 = +1 − − − (5) feature extraction, noise filtering, dimensionality
𝑠
reduction, and visualization (Liakos et al., 2018).
3.4 Framework overview For instance, the first 49 principal components
keep 80% of the overall variance in grayscale
The process begins by labelling and
images from the Maize dataset, which have
inputting plant leaf images into a shallow CNN
256x256 pixel dimensions (a total of 62,350
derived from the pre-trained VGG-16 model
pixels). The first 249 components preserve 90% of
through transfer learning, leveraging features from
the variation. This allows at least 90% of the image
the ImageNet dataset. This approach ensures
information to be preserved while representing the
efficient feature extraction while conserving
original 62,350 pixels with a 249-dimensional
computational resources. The extracted
vector called pixel features.[4]
embeddings undergo dimensionality reduction
using PCA to retain 99% variance, reducing
calculation costs and mitigating overfitting risks. 3.4.3 Image dataset
Finally, the embeddings are processed by Overall more than 3000 images were
classical classifiers like Kernel SVM and Random selected for the work. They are categorized into 3
Forest, termed SCNN-KSVM and SCNN-RF based on the features of the images. With 80% and
respectively, to evaluate the shallow CNN's 20% of the total data, respectively, the collection
performance across different classifiers[4]. has been divided into training and test sets[5].
Segmentation and classification are the two
3.4.1 Shallow CNN primary sequential stages of the PLDPNet system,
In recent plant disease identification which provides an end-to-end method for illness
studies, popular deep CNN models like VGG-16, prediction[6]. The CNN model VGG net placed
VGG-19, Inception-V3, and Xception have been second in the competition with a top-5 error rate of
widely adopted. VGG-16, known for its relative 7.5% on the validation set[9].
simplicity with 16 convolutional and fully connected
layers, still consists of approximately 138 million
parameters. However, this study focuses solely on
3.4.4 Multi-Class classification Methods Sensitivity Specificity Accuracy
Every sample in the dataset has a class ANN .96 .99 .99
label, and classification is the process of grouping KNN .94 .97 .96
SVM .92 .96 .94
records into particular classes based on feature
Naïve
values. Numerous deep learning (DL) and Bayes
.89 .95 .93
machine learning (ML) methods have been
created to predict classes for the test data and train
on the dataset.[7]. Comparison of Accuracy,
3.4.5 C-GAN model as synthetic image Specificity and Sensitivity
generator 1
A Conditional Generative Adversarial
0.9
Network (C-GAN) (Mirza and Osindero, 2014) can
be used as a data augmentation strategy to 0.8
increase the size of the dataset in order to reduce WOA-ANN KNN SVM Naïve Bayes
overfitting. In GANs, an image matrix is created
Sensitivity Specificity Accuracy
from random noise using conventional
convolutional layers. Two models are part of the
GAN architecture: a discriminator and a Fig. 4: The comparative performance of various models
generator.[15]. or methodologies in terms of three critical evaluation
metrics: accuracy, specificity, and sensitivity. The
graphical depiction allows for an intuitive understanding
4. RESULT AND DISCUSSION of the strengths and weaknesses of each approach,
aiding in performance analysis and model selection .
Using MATLAB, the enhanced deep
learning system divided the database into two Table 4: comparison of the precision, recall, and F-
primary groups: 80% for training and 20% for measure values for various models or methodologies,
evaluation. The first class was used for training, offering a comprehensive evaluation of their performance
in classification tasks.
while the second category was used for the whole
evaluation test.
Methods Precision Recall F-Measure
Estimating the degree of damage to olive
leaves and its percentage to the total leaf area was ANN .97 .98 .98
made easier by the forecasting of olive diseases
KNN .94 .94 .94
using the optimized artificial neural networks
technique. By separating the peacock disease SVM .91 .91 .91
area from the remaining leaf area, the method
increases the precision of picture evaluation and
categorization.Healthy leaves, which are free of Naïve
.89 .89 .89
any blemishes, are easily identifiable. Bayes

The study contrasts the effectiveness of

the suggested strategy with other classification
techniques. The specificity, sensitivity, and
Comparison of Precision,
accuracy values are shown in Table 3, illustrates Recall, and F-measure
these measures schematically, demonstrating that 1
the optimized ANN surpasses SVM, K-nearest
0.95
neighbours, and Naive Bayes techniques. Table 3
compares F-measure, Precision, and Recall 0.9
values. 0.85
0.8
Table 3: Comparison of key performance metrics— WOA-ANN KNN SVM Naïve Bayes
accuracy, specificity, and sensitivity—across various
methodologies, models, or systems. Precision Recall F-Measure
Fig. 5: provides a visual representation of how different Alexandria Eng. J., vol. 72, pp. 213–224, 2023,
models or methodologies perform across the three doi: 10.1016/j.aej.2023.03.081.
critical metrics: precision, recall, and F-measure.

6. F. Arshad et al., “PLDPNet : End-to-end hybrid

5. CONCLUSION deep learning framework for potato leaf
Predicting plant diseases early is crucial for disease prediction,” Alexandria Eng. J., vol. 78,
safeguarding crops and improving living standards no. June, pp. 406–418, 2023, doi:
globally. A significant challenge that needs 10.1016/j.aej.2023.07.076.
addressing to promote economic growth and
enhance agricultural output is the timely detection 7. R. Sujatha, J. Moy, N. Z. Jhanjhi, and S.
of plant diseases. Conventional techniques for Nawaz, “Microprocessors and Microsystems
detecting diseases are expensive and time- Performance of deep learning vs machine
consuming. The automatic diagnosis of plant learning in plant leaf disease detection,”
diseases through the use of deep learning (DL) Microprocess. Microsyst., vol. 80, no.
and machine learning (ML) is a significant November 2020, p. 103615, 2021, doi:
development in early disease detection. The 10.1016/j.micpro.2020.103615.
significance of domain-specific initiatives that set
standards for ML/DL applications and draw 8. M. Badiger and J. Alex, “Tomato plant leaf
attention to the unrealized potential of ML and DL disease segmentation and multiclass disease
in this field is highlighted by this analysis. An detection using hybrid optimization enabled
optimized artificial neural network designed deep learning,” J. Biotechnol., vol. 374, no.
especially for the detection and classification of August, pp. 101–113, 2023, doi:
plant diseases is presented in this paper. The 10.1016/j.jbiotec.2023.07.011.
algorithm incorporates feature extraction, feature
selection, and classification procedures after the 9. E. Chebet, L. Yujian, S. Njuki, and L. Yingchun,
“A comparative study of fi ne-tuning deep
pre-processing phase.
learning models for plant disease identi fi
cation,” Comput. Electron. Agric., vol. 161, no.
REFERENCE February 2018, pp. 272–279, 2019, doi:
10.1016/j.compag.2018.03.032.
1. P. Sajitha, A. D. Andrushia, N. Anand, and M.
Z. Naser, “A review on machine learning and 10. “Plant Disorders”, [Online]. Available:
deep learning image-based plant disease https://daneshyari.com/
classification for industrial farming systems,” J.
Ind. Inf. Integr., vol. 38, no. January, p. 100572, 11. J. Garcia and A. Barbedo, “Impact of dataset
2024, doi: 10.1016/j.jii.2024.100572. size and variety on the e ff ectiveness of deep
learning and transfer learning for plant disease
2. U. Sanath Rao et al., “Deep Learning Precision classi fi cation,” Comput. Electron. Agric., vol.
Farming: Grapes and Mango Leaf Disease 153, no. July, pp. 46–53, 2018, doi:
Detection by Transfer Learning,” Global 10.1016/j.compag.2018.08.013.
Transitions Proceedings, vol. 2, no. 2. pp. 535–
544, 2021. doi: 10.1016/j.gltp.2021.08.002. 12. R. Dogra, S. Rani, A. Singh, M. A. Albahar, A.
E. Barrera, and A. Alkhayyat, “Deep learning
3. D. Aditya, R. G. Manvitha, and C. L. R. Mouli, model for detection of brown spot rice leaf
“Detect-o-thon : Identification of infected plants disease with smart agriculture,” Comput.
by using deep learning,” Glob. Transitions Electr. Eng., vol. 109, no. PA, p. 108659, 2023,
Proc., vol. 2, no. 2, pp. 336–343, 2021, doi: doi: 10.1016/j.compeleceng.2023.108659
10.1016/j.gltp.2021.08.034.
13. P. Kumar Yadav et al., “Detecting volunteer
4. Y. Li, J. Nie, and X. Chao, “Do we really need cotton plants in a corn field with deep learning
deep CNN for plant diseases identification ?,” on UAV remote-sensing imagery,” Comput.
Comput. Electron. Agric., vol. 178, no. October, Electron. Agric., vol. 204, no. July 2022, p.
p. 105803, 2020, doi: 107551, 2023, doi:
10.1016/j.compag.2020.105803. 10.1016/j.compag.2022.107551
5. H. H. Alshammari, A. I. Taloba, and O. R. 14. S. Verma, P. Kumar, and J. P. Singh, “Original
Shahin, “Identification of olive leaf disease papers A meta-learning framework for
through optimized deep learning approach,”
 recommending CNN models for plant disease on NPK Levels using Machine Learning
identification tasks,” Comput. Electron. Agric., Algorithms," 2024 International Conference on
vol. 207, no. February, p. 107708, 2023, doi: Computing and Data Science (ICCDS),
10.1016/j.compag.2023.107708. Chennai, India, 2024, pp. 1-6, doi:
10.1109/ICCDS60734.2024.10560386.
15. A. Abbas, S. Jain, M. Gour, and S.
Vankudothu, “Tomato plant disease detection 17. B Vimaladevi, M , Kiruba, “Farm Monitoring
using transfer learning with C-GAN synthetic and Controlling Based On Precision
images,” Comput. Electron. Agric., vol. 187, no. Agriculture”, International Journal of Recent
April, p. 106279, 2021, doi: Engineering Research and
10.1016/j.compag.2021.106279. Development(IJRERD), ISSN:2455:8761
Volume 3, Issue 4, 2018, Pages 25-30.
16. M. Vimaladevi, T. R, D. K, K. G and A. S,
"FertiForecast: Identification of Fertilizer Based